monk

a razor-thin QOL wrapper around OpenAI's Responses API

usage

1. import and create

import Monk from 'monkai';
const monk = new Monk('gpt-5');

2. add tools

monk.toolbox.create({
    name: 'say_hi',
    description: 'Say hi!',
    parameters: {
        'type': 'object',
        properties: {
            name: { 'type': 'string' },
        },
    },
    callback: ({ name }) => {
        if (name) {
            console.log(`Hi ${name}!`);
        } else {
            console.log('Hi!');
        }

        return { status: 'success' };
    },
});
monk.toolbox.create({
    name: 'get_date',
    description: 'Get the current date and time.',
    callback: () => new Date().toString(),
});

3. add messages and get responses

monk.addMessage(Monk.DEVELOPER, 'You are a helpful and friendly AI assistant.');
const completion = await monk.respond('Hello! My name is Hayden. What is the date tomorrow?');

... or equivalently ...

monk.addMessage(Monk.DEVELOPER, 'You are a helpful and friendly AI assistant.');
monk.addMessage(Monk.USER, 'What is the date tomorrow?');
const completion = await monk.respond();

complete API

• constructor(model: string)
• .client: OpenAI: underlying OpenAI client
• .model: string: model string provided in constructor
• .messages: object[]: all input messages, including sent messages as well as queued messages
• .toolbox: ToolBox
  • .create(options: object): create a function tool. options:
    • name: string (required)
    • description: string (required)
    • callback: (args: object) => any (required)
    • parameters: object
    • strict: boolean
  • .call(toolCall: object): call a function tool. toolCall:
    • name: string (required)
    • arguments: object (required)
• .addMessage(role: Monk.USER | Monk.DEVELOPER | Monk.ASSISTANT, content: string): add a message to the model input without sending it to the model
• async .respond(prompt: string | null): append prompt to input messages if provided, then get a response from OpenAI, calling tools automatically